Comparison of Normalization Methods for cDNA Microarrays

In a study done by Pritchard et al. [2001], normal gene expression variation was examined in six genetically identical male mice, to determine a baseline variation for gene expression studies in mice. In this paper, we use data from their study to accomplish the following three goals: 1) Evaluate five data normalization procedures along with two methods omitting data normalization, and study their impact on identifying baseline differentially expressed genes; 2) Perform pair-wise comparisons using McNemar's tests on five normalization methods and two methods omitting the normalization step; 3) Address data quality issues and examine the effect of normalization on analysis results for genes that do not meet either or both of the two data quality criteria. Depending on which normalization method is used, whether omitting the normalization step or not, the number of genes and the set of the genes identified as differentially expressed from the same study can be substantially different. Analysis demonstrates that when data quality is not ensured, performing normalization can add noise to the data and can bias gene-based ANOVA results. Thus we conclude that ensuring data quality and establishing quality control measures is crucial to increase the effectiveness of normalization procedures and the accuracy of data analysis results. The study also reconfirmed that proper experimental design and establishing rigorous data quality control standards are indispensable factors for the success of a microarray experiment.